Content & Media Creation · prime-skills/runcomfy-agent-skills

image-to-video

Name: image-to-video
Author: prime-skills/runcomfy-agent-skills

Animate any still image on RunComfy — this skill is a smart router that matches the user's intent to the right i2v model in the RunComfy catalog. Picks HappyHorse 1.0 I2V (Arena #1, native audio, identity preservation) for general animations, Wan 2.7 with `audio_url` for custom-voiceover lip-sync, or Seedance 2.0 Pro for multi-modal animation from image + reference video + reference audio. Bundles each model's documented prompting patterns so the caller gets sharper output without burning iterations on the wrong model. Calls `runcomfy run <vendor>/<model>/image-to-video` (or endpoint variant) through the local RunComfy CLI. Triggers on "image to video", "image-to-video", "i2v", "animate image", "make this move", or any explicit ask to turn a still into video.

L4 High-Privilege · Graded auth & full auditV1 Commit-PinnedExecute CommandNetwork AccessWrite FileDual-pass Audit

Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

i2v intent routing: Defaults to HappyHorse 1.0 I2V for general animation (portraits/products), routes to Wan 2.7 t2v+audio_url for custom voiceover lip-sync, and Seedance 2.0 Pro for multimodal synthesis; the agent classifies according to the intent table and lands on the corresponding subsection.

image-to-video/SKILL.md

| Animate a portrait — keep identity stable | **HappyHorse 1.0 I2V** | #1 on Artificial Analysis Arena (Elo 1392); strong facial fidelity |

The routing table contains 8 intent rows + a default row (Default if unspecified → HappyHorse), with the transition sentence `The agent reads this table, classifies the user's intent, and picks the matching subsection below.` (same writing template as video-edit) — classification is executed by the host LLM, with no code in the skill itself.

Honest flagging of endpoint gotchas: Wan 2.7 does not have an independent i2v endpoint under this catalog; lip-sync actually reuses the t2v endpoint + audio_url, explicitly declared in documentation to prevent the agent from blindly calling /image-to-video.

image-to-video/SKILL.md

**Model**: `wan-ai/wan-2-7/text-to-video` (NOT `/image-to-video` — Wan 2.7's t2v endpoint accepts an `audio_url` that drives lip-sync)

A note in the same section adds: `For pure i2v (image animated by motion prompt only), prefer HappyHorse i2v.` — demonstrating the schema is organized from empirical testing against the RunComfy catalog rather than assumptions; description also reserves the '(or endpoint variant)' phrasing in alignment with this.

Multimodal synthesis reference protocol: Seedance 2.0 Pro accepts arrays of 0–9 images, 0–3 reference videos, and 0–3 reference audios, routing multi-modal cues to the model using prompts numbered as 'image 1 / video 1 / audio 1'.

image-to-video/SKILL.md

- **Number the refs** in the prompt: `"subject from image 1, lighting from video 1, voice from audio 1"`. Seedance routes cues correctly.

The schema row provides array limits (`image_url` array 0–9, `video_url` 0–3, `audio_url` 0–3), and prompt discipline guidelines emphasize the first image as the primary subject (`**First is the primary subject.**`).

Motion prompt syntax: Front-loading action verbs, avoiding restating image contents, and keeping a single primary motion per clip, standardizing the phrasing for 'bringing static images to life' into a template.

image-to-video/SKILL.md

- **Lead with motion verbs**: "drift", "dolly in", "orbit", "tilt up", "reveal", "blink", "breathe". Front-load what's MOVING.

Coupled with guidelines `Don't restate the image — the model sees it. Focus tokens on what changes.` and `One beat per clip — single primary motion` — addressing the common i2v pitfall where verbose descriptions dilute motion.

Delegates execution to the runcomfy CLI + cloud inference: After selecting one of the three paths, invokes `runcomfy run <model_id>` with a matching JSON body, while the CLI handles submission, polling, and downloading artifacts.

image-to-video/SKILL.md

The skill picks one of HappyHorse 1.0 I2V / Wan 2.7 t2v+audio / Seedance 2.0 Pro based on user intent and invokes `runcomfy run <model_id>` with the matching JSON body.

Heavy video generation workloads remain entirely in the cloud; the skill itself serves as a documentation carrier for 'routing decisions + parameter templates + prompt disciplines'.

2Core Capabilities

01Three-way i2v intent routing (portrait/product general animation, custom voiceover lip-sync, multimodal synthesis)

02Custom voiceover lip-sync driven by audio_url (reusing Wan 2.7 t2v endpoint)

03Same-image multilingual dubbing variants: lock seed and swap audio_url to generate video per language

04Multimodal reference synthesis (arrays of images + reference videos + reference audios, with the first image as primary subject)

05Motion/camera/lighting prompt syntax + negative_prompt exclusions

06Parameterized generation: per-field control over resolution, duration, aspect ratio, seed, and watermark toggle

3External Dependencies

Kind	Dependency / Endpoint
cli	runcomfy（@runcomfy/cli 官方 CLI）
package	@runcomfy/cli（npm 全局安装）
network	RunComfy Model API（请求提交）
network	生成结果下载白名单
network	云端模型路由（经 CLI/Model API 访问）：happyhorse/happyhorse-1-0/image-to-video、wan-ai/wan-2-7/text-to-video、bytedance/seedance-v2/pro
network	RunComfy 模型页 / 文档站（引用链接）
network	用户提供的源图/参考视频/参考音频 URL——由 RunComfy 模型服务器抓取

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Graded/per-use authorization, sandbox recommended

For Administrators

Approval workflow, rollback support, full audit

Evidence & Triggers (File Paths & Excerpts)

网络端点: Declares external network access: model-api.runcomfy.net — i2v request submission (the only submission channel declared in documentation, via runcomfy CLI)
本地工作目录: File I/O and artifact output: CLI downloads generated videos to a local directory specified by --output-dir (path determined by caller)
SKILL.md: Declares calls to host CLI interpreter or external execution environment

Audit Notes: Per unified grading standards: Runtime invokes third-party npm package/CLI @runcomfy/cli, which reads user RunComfy credentials (~/.config/runcomfy/token.json or RUNCOMFY_TOKEN env var) on every invocation — triggers 'invokes 3rd-party npm/CLI and potentially reads quota/keys → orange'; execution surface (CLI, Model API, token reading) is entirely outside the repo, and doc claims (0600, 2 GiB limit, no shell expansion) cannot be verified in this pin, following 'when in doubt, take the higher tier' to assign orange. Outbound network traffic is also present: source images/reference audio-video (including voiceover tracks containing voice data) and prompts are sent to third-party cloud via model-api.runcomfy.net; recipient is predictable but assets leave boundary; reference media URLs are fetched server-side. Skill itself is pure markdown, no scripts, no obfuscation, no hidden endpoints, avoiding red tier.

5Second-Pass Verification

[ok] 三个 i2v 模型 ID 与路由表、调用示例一致 — happyhorse/happyhorse-1-0/image-to-video、wan-ai/wan-2-7/text-to-video（非 image-to-video）、bytedance/seedance-v2/pro 三处出现位置均成对自洽。
[unlocatable] HappyHorse『Arena #1 (Elo 1392)』宣称 — 路由表 Why 列与 Route 1 标题（'Arena rank: #1 (Elo 1392)'）均引用站外 Artificial Analysis 排名，pin 内无任何数据支撑；无法复核其时效与真伪，标记待补。
[ok] Wan 2.7 走 t2v 端点做口型同步的端点异例 — Route 2 明确 'NOT /image-to-video'，与 description 的 '(or endpoint variant)' 措辞呼应；无自相矛盾。
[ok] 限制声明一致性（Seedance 720p ceiling、时长上限、音频 2–15s/<15MB） — Limitations 节与三处 schema 表（duration/resolution/audio 规格）数字一致。
[unlocatable] SKILL.md 对 CLI 行为的断言（无 shell 展开、token 0600、2 GiB 上限） — CLI 源码不在本仓库，仅文档自述，无法独立证实。
[discrepancy] 身份/安装链接一致性 — SKILL.md 安装与 GitHub 链接指向 agentspace-so/runcomfy-skills，镜像 README 指向 agentspace-so/runcomfy-agent-skills，被 pin 仓库为 prime-skills/runcomfy-agent-skills——三处来源标识不一致（发布身份漂移，非代码缺陷）。

6Verdict & Scenarios

Splits intents into three paths by use case (general animation / voiceover lip-sync / multimodal) with clear defaults, preventing wasted iterations on the wrong model

Explicitly flags endpoint pitfalls (Wan uses t2v+audio_url), making it more reliable than docs that blindly call an i2v endpoint just because the skill is named i2v

Clear multimodal reference protocol: array limits + numbered cue syntax to combine image/video/audio references across three channels

Actionable and concrete motion prompt disciplines (front-loading verbs, no image restatement, single primary motion) targeting common i2v drift issues

Transparent self-reported security posture, including notices on third-party media fetching and image prompt injection risks

Best for: Ideal for agent users with a RunComfy account needing to animate static/product images, generate custom voiceover lip-sync, or produce multilingual dubbed assets; batch scenarios such as e-commerce product animation, marketing shorts, and multilingual spokesperson videos; also fits skill authors looking to adopt the pattern of 'honestly documenting endpoint anomalies'.

Not for: Not suitable for users without a RunComfy account or unwilling to upload assets to the cloud; scenarios requiring text-to-video from scratch without source images (this skill is an i2v router); scenarios needing to blend capabilities of multiple models in a single clip (e.g., HappyHorse animation + Wan lip-sync) — documentation clearly states separate calls and manual stitching are required; not for users seeking local offline generation or end-to-end auditable setups.

Install Agent-Ready

① Verified Mirror 2026-09-09

Option A · Direct DownloadDownload image-to-video.tar.gz

sha256: 72d420bb6e15ce52…

Option B · Copy Guide for Agent

install.json

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit fca19ae084.